What a free check looks like
This is the whole thing. Nothing held back except the analysis you would be paying for.
A real reading, run on 10 August 2026. COSRX has not asked for this and is not a client. We measured a well-known brand so you can check the result yourself.
Prepared for
COSRX
cosrx.com
Singapore
10 August 2026
2 of 6
Named in 2 of the 6 questions that did not mention COSRX. Those are the ones that decide whether a new customer finds you.
Across all ten questions the figure is 6 of 10, but four of those questions named COSRX in the question itself, so being mentioned back proves nothing. We report the honest number first.
Who gets recommended instead
Every brand named across the six discovery questions.
Also named once each: ETUDE SoonJung, Dr.G, ILLIYOON, AXIS-Y, Haruharu Wonder.
What we found
Three of the six discovery answers cited a brand's own website : Anua, Biodance, Beauty of Joseon. In two of those three the brand that owned the cited page was also named in the answer. Biodance was the exception, cited but not named, which is the caveat worth carrying: being the source an answer reads is not the same as being the brand it recommends.
COSRX's own site was cited exactly once, on the question that already said the word "COSRX". Its pages are not being pulled into the questions where a new customer is deciding. That is the gap, and it is a content problem rather than a product problem.
The limit of what we can promise
Three questions came back with no citations at all (both head-to-head comparisons and the reputation question).
Those answers came from what the model already holds, not from anything it went and read. Publishing will not move them in the short term, and anyone who tells you otherwise is guessing. Only the category and constraint questions had live retrieval, so those are the ones worth acting on.
The questions we asked
Each asked in its own fresh, memory-off session, so no answer could influence the next. Sources are the domains that answer actually cited.
| Question | COSRX | Cited |
|---|---|---|
| best korean skincare brand for oily acne-prone skin | Named | anua.com, allure.com |
| best korean moisturiser for dry skin | Not named | allure.com |
| best korean sunscreen for daily use in humid weather | Not named | harpersbazaar.com.sg, reddit.com |
| which korean skincare brands are good for sensitive skin | Not named | vogue.com, glow.com.au, reddit.com |
| best korean skincare under $20 that actually works | Named | biodance.com, vogue.com, glow.com.au |
| fragrance free korean skincare for reactive skin | Not named | beautyofjoseon.com, k-beautyskin.com |
| Questions that named COSRX in the question | ||
| cosrx or beauty of joseon, which is better for acne | Chose COSRX | none |
| is anua heartleaf toner better than cosrx snail mucin essence | Chose COSRX | none |
| is cosrx a good skincare brand | Positive | none |
| what are the best cosrx products to start with | Named | cosrx.com |
How this was measured
| Assistant | ChatGPT, on its standard consumer web app, not a developer API. Your own report names the assistant behind every question, and on a paid plan covers four or seven of them. |
| Questions | 10, each in its own session |
| Session state | Memory off, no chat history, no custom instructions |
| Market | Singapore. The assistant returned SGD pricing and local retailers unprompted |
| Date | 10 August 2026 |
| Evidence | Every answer recorded verbatim with its cited URLs |
Memory off matters more than it sounds. Asked the same question on a logged-in account with memory on, the same assistant put a different brand first and cited a different set of sources. A reading taken in your own account tells you what AI says to you, not what it says to a stranger deciding what to buy.
AI answers vary between sessions and change over time. This is a reading taken on one date, not a permanent score.
What the full report adds
75 questions instead of 10. Four assistants instead of one. Every source deciding who gets named, tracked month over month, and a ranked list of what to publish.
Both are published in full on a real brand we have never worked for: the free check on COSRX and the monthly report that followed it. Read them before deciding whether any of this is worth your time.